CN111339941A - Head posture detection method - Google Patents
Head posture detection method Download PDFInfo
- Publication number
- CN111339941A CN111339941A CN202010119229.2A CN202010119229A CN111339941A CN 111339941 A CN111339941 A CN 111339941A CN 202010119229 A CN202010119229 A CN 202010119229A CN 111339941 A CN111339941 A CN 111339941A
- Authority
- CN
- China
- Prior art keywords
- axis
- head
- detection method
- deep learning
- learning model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 21
- 238000013136 deep learning model Methods 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000000034 method Methods 0.000 claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000013507 mapping Methods 0.000 claims abstract description 4
- 238000012360 testing method Methods 0.000 claims abstract description 4
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 206010003805 Autism Diseases 0.000 description 3
- 208000020706 Autistic disease Diseases 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 208000013403 hyperactivity Diseases 0.000 description 3
- 208000020016 psychiatric disease Diseases 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 208000016686 tic disease Diseases 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000001755 vocal effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/242—Aligning, centring, orientation detection or correction of the image by image rotation, e.g. by 90 degrees
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a head posture detection method, which comprises the following steps: (a) selecting a data set; (b) preprocessing the face pictures in the data set, and then converting the sizes of the face pictures to obtain pictures with set sizes; (c) constructing a deep learning model by taking MobileNetv2 as a backbone; (d) putting the pictures with the set sizes into the neural network for classification; (e) applying softmax to the fully-connected layer result to map the fully-connected layer value to a probability value; (f) mapping the probability value to obtain regression, and calculating the regression loss probability by using an MSE loss function method; (g) carrying out weight weighted summation on the loss probability, and carrying out gradient direction on the final loss probability to finish the training of a deep learning model; (h) and testing the head of the child by using the deep learning model. The method has the advantages of high program detection speed and real-time performance.
Description
Technical Field
The invention relates to a head posture detection method, in particular to a method for detecting the head posture of a child with mental diseases by utilizing a deep learning training model in a computer vision technology.
Background
Head gestures can help people position to convey some rich information, such as people pointing with their head to indicate their dialog objects and intentions. In some dialogs, the head direction is a non-verbal announcement that alerts the listener when to change roles and begin speaking; in these dialogs, the head pose direction and the form of the gesture have the same important role.
For some children with autism, hyperactivity or tic disorder, the head orientation can reflect what the children aim at the current environment, so that the therapist or doctor can know the thoughts of the children conveniently. There are several methods of detecting head pose today: such as the early use of the detector array approach (training a large number of head detectors, each detector accommodating a particular pose, then assigning a discrete pose to these detectors, with some head pose being predicted accordingly); in the middle period, a nonlinear regression method or a random forest algorithm in machine learning is used; some recent algorithms extract key points of the human face to perform prediction of the head pose by deep learning training.
However, the above method has certain disadvantages: are relatively dependent on environmental influences. If the environmental background is changed greatly or the ages of the testers are different greatly (for example, the people with mental diseases such as autism, hyperactivity or tics are children generally, but the head posture of the children is detected slightly different from that of adults), the detection structure is easy to be inaccurate.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a head posture detection method which is suitable for children with mental diseases such as autism, hyperactivity or tic disorder.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows: a head posture detection method comprises the following steps:
(a) selecting a data set;
(b) preprocessing the face picture in the data set, performing face detection and cutting on the face picture by using a multitask cascade convolution neural network, and then performing size conversion to obtain a picture with a set size;
(c) constructing a neural network which takes MobileNetv2 as a backbone and is respectively connected with three full-connection layers for the deep learning model;
(d) putting the pictures with the set sizes into the neural network for classification;
(e) applying softmax to the fully-connected layer result to map the fully-connected layer value to a probability value;
(f) mapping the probability value to obtain regression, and calculating the regression loss probability by using an MSE loss function method;
(g) carrying out weight weighted summation on the loss probability, and carrying out gradient direction on the final loss probability to finish the training of a deep learning model;
(h) the nose is taken as a basic point, the horizontal direction is set as an x axis, the vertical direction is set as a y axis, the z axis is perpendicular to a plane formed by the x axis and the y axis, the angle of clockwise rotation around the x axis, the y axis and the z axis is defined as the offset angle of the head posture in pitch, yaw and roll directions, and the deep learning model is used for testing the head of the child to obtain the posture position of the head of the child.
Optimally, in step (a), the data sets are BIWI, 300W-LP and AFLW2000 data sets.
Optimally, in the step (b), the preprocessing is to exclude the unwanted background or other objects in the face picture.
Further, in step (b), the multitask cascade convolution neural network is completed by three cascade lightweight CNNs of PNet, RNet and Onet.
Optimally, in the step (d), the classification result map is put into a range.
Due to the application of the technical scheme, compared with the prior art, the invention has the following advantages: according to the head posture detection method, the deep learning model is used for performing loss calculation on the three angles by using the composite loss function, the program detection speed is high, and the real-time performance can be achieved; the unified evaluation standard is available, and the accuracy is high; the time for the therapist or doctor to observe the child can be saved, and the therapist or doctor can treat the child more in other aspects; and the data may be stored and displayed in the form of video.
Drawings
FIG. 1 is a flow chart of an MSE loss function in the head pose detection method of the present invention;
FIG. 2 is a diagram illustrating a first effect of the head pose detection method according to the present invention;
fig. 3 is a diagram illustrating a second effect of the head pose detection method according to the present invention.
Detailed Description
Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
The head posture detection method comprises the following steps:
(a) selecting a data set; in the present embodiment, the datasets are primarily BIWI, 300W-LP, and AFLW2000 datasets (i.e., primarily trained and tested on the BIWI, 300W-LP, and AFLW2000 datasets). The BIWI data set was released in 2010 and contained 1000 high quality 3D scanners and 3D data collected by professional microphones, with intensive dynamic face scans acquired at 25 frames per second. The 300W-LP is a 3D data set obtained by simulation based on a 300W data set and a 3DMM model, which is the most widely used simulation data set in the 3D field and comprises labels of 68 key points, camera parameters and coefficients of the 3DMM model. AFLW is a large-scale database of faces including multi-pose, multi-view faces, typically used to evaluate the effect of facial keypoint detection, with pictures from a crawl of flickr, totaling 21997 pictures, 25993 faces, and 21 keypoints per face label (380000 keypoints total).
(b) The method comprises the steps of preprocessing face pictures in a data set, using a multi-task cascade convolution neural network (MTCNN, MTCNN is a more classical and rapid face detection technology, and is completed by three cascade lightweight CNNs, PNet, RNet and Onet) to detect and cut the faces of the face pictures, and eliminating some unneeded backgrounds or other objects to ensure that overfitting data does not appear during training;
(c) constructing a neural network which takes MobileNetv2 as a backbone and is respectively connected with three full-connection layers for the deep learning model (namely, the deep network takes MobileNetv2 as the backbone of the foundation and is respectively connected with the three full-connection layers, and each layer is independently predicted);
(d) the pictures with set sizes are put into a neural network for classification, and then the classification result map is put into a range, so that the precision of the method is greatly improved (the step is the loss probability of classification);
(e) performing softmax on the result of the full link layer to map the value of the full link layer to a probability value;
(f) the probability value is mapped to obtain regression (namely, the regression is needed when mapping is carried out according to the probability value), and the loss probability of the regression is calculated by using an MSE loss function method (as shown in figure 1, MSE mainly introduces the definition of common loss function MSE in machine learning and derivation characteristics thereof, and the mean square error in mathematical statistics refers to the expectation value of the square of the difference between a parameter estimation value and a parameter value;
(g) carrying out weight weighted summation on the loss probability, and carrying out gradient direction on the final loss probability to finish the training of a deep learning model;
(h) taking a nose as a base point, setting the horizontal direction as an x-axis, setting the vertical direction as a y-axis, setting the z-axis to be perpendicular to a plane formed by the x-axis and the y-axis, defining an angle of clockwise rotation around the x-axis, the y-axis and the z-axis as a deviation angle of the head posture in pitch, yaw and roll directions, and testing the head of the child by using the deep learning model to obtain the posture position of the head of the child (for specific application, see fig. 2 and fig. 3).
The above embodiments are merely illustrative of the technical ideas and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the contents of the present invention and implement the present invention, and not to limit the protection scope of the present invention. All equivalent changes and modifications made according to the spirit of the present invention should be covered within the protection scope of the present invention.
Claims (5)
1. A head posture detection method is characterized by comprising the following steps:
(a) selecting a data set;
(b) preprocessing the face picture in the data set, performing face detection and cutting on the face picture by using a multitask cascade convolution neural network, and then performing size conversion to obtain a picture with a set size;
(c) constructing a neural network which takes MobileNetv2 as a backbone and is respectively connected with three full-connection layers for the deep learning model;
(d) putting the pictures with the set sizes into the neural network for classification;
(e) applying softmax to the fully-connected layer result to map the fully-connected layer value to a probability value;
(f) mapping the probability value to obtain regression, and calculating the regression loss probability by using an MSE loss function method;
(g) carrying out weight weighted summation on the loss probability, and carrying out gradient direction on the final loss probability to finish the training of a deep learning model;
(h) the nose is taken as a basic point, the horizontal direction is set as an x axis, the vertical direction is set as a y axis, the z axis is perpendicular to a plane formed by the x axis and the y axis, the angle of clockwise rotation around the x axis, the y axis and the z axis is defined as the offset angle of the head posture in pitch, yaw and roll directions, and the deep learning model is used for testing the head of the child to obtain the posture position of the head of the child.
2. The head posture detection method according to claim 1, characterized in that: in step (a), the data sets are BIWI, 300W-LP and AFLW2000 data sets.
3. The head posture detection method according to claim 1, characterized in that: in the step (b), the preprocessing is to exclude the background or other objects which are not needed in the face picture.
4. The head pose detection method according to claim 1 or 3, characterized in that: in the step (b), the multitask cascade convolution neural network is completed by three cascade lightweight CNNs of PNet, RNet and Onet.
5. The head posture detection method according to claim 1, characterized in that: in the step (d), the classification result map is put into a range.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010119229.2A CN111339941A (en) | 2020-02-26 | 2020-02-26 | Head posture detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010119229.2A CN111339941A (en) | 2020-02-26 | 2020-02-26 | Head posture detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111339941A true CN111339941A (en) | 2020-06-26 |
Family
ID=71183659
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010119229.2A Pending CN111339941A (en) | 2020-02-26 | 2020-02-26 | Head posture detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111339941A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112241761A (en) * | 2020-10-15 | 2021-01-19 | 北京字跳网络技术有限公司 | Model training method and device and electronic equipment |
CN112634363A (en) * | 2020-12-10 | 2021-04-09 | 上海零眸智能科技有限公司 | Shelf attitude estimation method |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748858A (en) * | 2017-06-15 | 2018-03-02 | 华南理工大学 | A kind of multi-pose eye locating method based on concatenated convolutional neutral net |
-
2020
- 2020-02-26 CN CN202010119229.2A patent/CN111339941A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748858A (en) * | 2017-06-15 | 2018-03-02 | 华南理工大学 | A kind of multi-pose eye locating method based on concatenated convolutional neutral net |
Non-Patent Citations (1)
Title |
---|
XIAOJIE GUO等: "PFLD: A Practical Facial Landmark Detector" * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112241761A (en) * | 2020-10-15 | 2021-01-19 | 北京字跳网络技术有限公司 | Model training method and device and electronic equipment |
CN112241761B (en) * | 2020-10-15 | 2024-03-26 | 北京字跳网络技术有限公司 | Model training method and device and electronic equipment |
CN112634363A (en) * | 2020-12-10 | 2021-04-09 | 上海零眸智能科技有限公司 | Shelf attitude estimation method |
CN112634363B (en) * | 2020-12-10 | 2023-10-03 | 上海零眸智能科技有限公司 | Goods shelf posture estimating method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109325437B (en) | Image processing method, device and system | |
CN110287880A (en) | A kind of attitude robust face identification method based on deep learning | |
CN111028319B (en) | Three-dimensional non-photorealistic expression generation method based on facial motion unit | |
Li et al. | Sign language recognition based on computer vision | |
CN110889343A (en) | Crowd density estimation method and device based on attention type deep neural network | |
CN112818969A (en) | Knowledge distillation-based face pose estimation method and system | |
CN112949622A (en) | Bimodal character classification method and device fusing text and image | |
CN111339941A (en) | Head posture detection method | |
Depuru et al. | Convolutional neural network based human emotion recognition system: A deep learning approach | |
CN112906520A (en) | Gesture coding-based action recognition method and device | |
Yanmin et al. | Research on ear recognition based on SSD_MobileNet_v1 network | |
Zhao et al. | Rapid offline detection and 3D annotation of assembly elements in the augmented assembly | |
CN111914595A (en) | Human hand three-dimensional attitude estimation method and device based on color image | |
CN109753922A (en) | Anthropomorphic robot expression recognition method based on dense convolutional neural networks | |
CN110490165B (en) | Dynamic gesture tracking method based on convolutional neural network | |
Wang et al. | Swimmer’s posture recognition and correction method based on embedded depth image skeleton tracking | |
CN116823983A (en) | One-to-many style handwriting picture generation method based on style collection mechanism | |
CN115496859A (en) | Three-dimensional scene motion trend estimation method based on scattered point cloud cross attention learning | |
Chen et al. | Intelligent Recognition of Physical Education Teachers' Behaviors Using Kinect Sensors and Machine Learning. | |
CN113536926A (en) | Human body action recognition method based on distance vector and multi-angle self-adaptive network | |
Zhou et al. | Motion balance ability detection based on video analysis in virtual reality environment | |
Gai et al. | Digital Art Creation and Visual Communication Design Driven by Internet of Things Algorithm | |
Liang et al. | Interactive Experience Design of Traditional Dance in New Media Era Based on Action Detection | |
CN112989952B (en) | Crowd density estimation method and device based on mask guidance | |
Xu et al. | Research on computer graphics and image design and visual communication design |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200626 |